JP6404915B2 - Automatic data compression - Google Patents

Description

  This document relates to control units and systems that automatically determine whether data is transferred in an encoded manner, particularly in the context of distributed processing of applications.

  In recent years, data, computer programs and applications are increasingly stored and executed in network-centric data centers (“clouds”). However, especially in the technical field of distributed real-time processing of media data in the cloud, due to bandwidth and end-to-end delay constraints, processing of real-time media particularly requires sensitivity with respect to transmission data across network links. .

  In a service-type platform (PaaS) approach, a function is provided for automatically distributing an application composed of separable software components on processing nodes arranged in different parts of a network. . Utilize shared data locality to minimize traffic in the transmission network, optimize processing node resource utilization, and improve user experience by processing delay sensitive tasks close to the user's location in the network This distribution is used to do this. In a cloud-based processing environment, different data transmission scenarios between different software tasks executed on separate processing nodes can be applied. Typical scenarios are media data transmission to and from end devices to media processing nodes across the network, media data transmission between processing nodes within a single data center, and within different data centers. Media data transmission between arranged processing nodes.

  A typical application scenario that benefits from distributed processing on a service platform is, for example, video chat. In a video chat scenario, for example, a group of people participate in a virtual shared video environment / video chat room. Each individual streams their camera video data to the cloud. The video chat service combines all media streams of users gathering in such a shared environment and generates (ie, processes) a personal view of the virtual room for each participant. Thus, each individual in the virtual room receives a personalized rendered view of all (other) participants.

It is assumed that an application such as a video chat can be divided into separate processing parts (ie processing components) that are executed across different processing nodes in the cloud. A typical scenario includes the following:
-Camera signal preprocessing, eg background subtraction, face or action detection, eg “who is talking” function, which is preferably located on a processing resource located near the data source.
A module controlled by user interaction to determine a specific part of the video to be extracted from the real camera signal with high resolution by selecting the pan / tilt and zoom elements of the virtual camera.
A video mixing function that merges all of the different video signals of people in a chat room on a processing resource in the cloud, preferably placed in the “center of delay” for all participants.
-Personal viewpoint rendering, preferably located on processing resources located near this user, controlled by end-user interactive gestures or other means.

  For example, the processing portion within the video chat application can be background extraction. This part may be constructed and composed of several processing components that perform separate tasks such as color conversion, comparison of the actual frame with the background model, statistical update of the background model, and foreground mask generation.

  For example, in a platform scenario, the components can communicate via a message passing interface that connects the output port of the sending component to the input port of one or more receiving components. Developers can create and maintain running instances of their applications by organizing a dynamic service logic graph composed of different processing elements that make up the application.

  Manipulating the graph includes creating, modifying or removing components and establishing or releasing connections between components in a dynamic manner. From the developer's perspective, the platform behaves like a single virtual processing resource, and it is the platform's job to properly deploy the different components of the application onto the physical infrastructure resources as described above.

  Typically, the processing power required for a particular task depends on different media parameters. The parameters can be image resolution, frame rate, number of participants, and the like. Furthermore, the number of resources required to perform a desired function may depend on the resource capacity and the current load status of the individual processing nodes (which may vary quickly).

  Thus, it is generally difficult for application developers to predict how different components of an application will be distributed most efficiently across different processing nodes. Since only the platform has the necessary information, the resource bottleneck that appears when the service on the resource is executed is re-assigned to some other processing resource / component that is currently not being used. It is the platform's job to decide how to resolve it by distributing it.

  However, the fundamental problem for distributing components of a video application such as a video chat service is that many media processing components primarily operate on a series of pixel values (ie, a single video frame), so The amount of data exchanged at is large. Thus, the component typically sends a complete frame of pixel values, ie an uncompressed video frame (raw data), to its output port. This makes it difficult to divide application components across different processing nodes as raw “video” traffic needs to be transmitted or exchanged over networks that connect different processing nodes, and thus network congestion. May bring. Thus, media data exchanged between components executing on different processing resources needs to be compressed in a meaningful way.

  Developers in the service graph during development do not know how the platform will split different service components across processing resources during service execution, so compression (ie encoding) in the communication path as needed / Platform functions that automatically include

  This document addresses, among other things, the technical problem of providing a control unit, system and method usable in a system having at least two of said processing nodes, such as processing nodes connected to each other in a network-centric data center. To do. The control unit, system and method can automatically determine whether data encoding and decoding is necessary and dynamically instantiate a data dependent compression mechanism between the transmitting and receiving components.

  By analyzing the data format of the message, including the header and data to be transferred (payload) and passing through the external network interface, the aspects described herein insert a data format dependent encoder at the transmitter side. And sending a control message indicating the compression format (ie, data format) to the receiver, and dynamically compressing each subsequent message using an instance of the compressor. Furthermore, the data transferred in the message can be decrypted remotely and delivered to the receiving component after decryption.

  Note that all of this can be done transparently to developers of such services, because the service logic flow needs to be paid attention when the components are run on the same or separate resources Because there is no. Media compression is included in the component's communication path autonomously by the underlying platform and only when necessary.

  Note that the reverse is also covered. If two components that were previously operating on different resources are placed on the same resource at the time of service execution for some reason, the previously external communication path will be internal, so media encoding and decoding The function will be automatically removed from the communication path by the platform.

  Thus, the control units and systems presented herein do not allow components connected via a communication channel to be aware of the difference in whether data is received from a local component or a remote component, and at the same time Has the technical advantage that the data rate on the network link is reduced. Thus, the entire compression scheme is hidden from the developer, the developer pays attention only to the components, and the components are unaware that compression has occurred within their connections.

  Specifically, this document specifically detects data between transmitting and receiving components by detecting that the communication channel between components uses an external transmission network for some reason instead of the communication channel inside the node. A control unit, autonomous system and method for a network-centric data center, i.e. a cloud environment, is provided that can dynamically instantiate dependent compression mechanisms.

  According to one aspect, the control unit can be configured to determine whether a communication channel between two processing nodes crosses a processing node boundary. When the communication channel crosses the boundary of one of the processing nodes, the control unit can be further configured to determine whether encoding / decoding of data transferred via the communication channel is necessary. . The control unit can be part of a processing node, for example a transmission processing node or a reception processing node. Alternatively or in addition, the control unit can be provided by a separate entity of the system with different processing nodes located in different data centers.

  According to a further aspect, a system is provided that can comprise at least two processing nodes connected by a network. Each node may comprise an execution environment and at least one processing component. Each execution environment may include means for connecting an outgoing port of a component transmitting data to an incoming port of a component receiving data via a communication channel. The system, preferably the control unit of the system, can be configured to determine whether the communication channel crosses a node boundary, as described above. Further, the system, preferably the control unit of the system, may be configured to determine if the data transferred via the communication channel needs to be encoded / decoded when the communication channel crosses a node boundary. it can.

  Node border crossings can be detected when the communication channel uses an external transmission network, such as a local area network (LAN) or a wide area network (WAN). For example, if the node is a unit comprising a processor, storage means and a network interface, the external transmission network can be defined as a network that connects the nodes, more specifically, the network interfaces of the nodes to each other. . Crossing a node boundary can be defined as crossing a network link. Preferably, node crossing can be detected when network protocols such as RTP, UDP, TCP, IP are used to send data over the communication channel. For example, a network boundary is traversed when one processing component operates on one network node and another processing component that communicates with the first component operates on another network node.

  Note that hereinafter the term “system” can be interchanged with the word “platform”. The term “data” can refer to information exchanged / transferred between parts / nodes / components of the system. The term “message” substantially means the header of the message and the data (payload) to be transmitted.

  The term “encoding / decoding” refers to encoding and / or decoding. Specifically, encoding and decoding means that the data is encoded (compressed) on one side from which the message / data is transmitted and the data is decoded on the other side or the receiving side. .

  In the following, it is assumed that the first component is the component that transmits data, i.e. the "sending component", and that the second component is the component that receives data, i.e. the "receiving component". Become.

  However, it should be noted that the system can comprise more than two nodes and more than two components. In addition, a plurality of different communication channels can be established between two or more nodes. In particular, the transmitting component can be configured to be connected to at least one receiving component via one or more communication channels.

  The terms “receiving node” and “transmitting node” refer to a node comprising receiving / transmitting components. The term node may refer to a computing resource or host that preferably comprises at least one processor, at least one storage device, and at least one network adapter. In other words, the node is preferably a fully operational stand-alone processing unit connected to other nodes in the network.

  According to a further aspect, the transmitting component can be configured to be connected to at least one receiving component via one or more communication channels. Application components can be flexibly distributed by establishing one or more communication channels between the component (s).

  Further, the system can comprise a table that includes information about each established communication channel. The information may include, for each communication channel, the ports of the transmission component and the reception component of that communication channel. Further, the information can include an identifier when the receiving component is located in a different node than the transmitting component. The identifier can identify the network location of the remote receiving node. The table can be located and stored in a specific (dedicated) processing node, and preferably a local copy is available at each processing node. In addition, each processing node may have a partial copy of the table that includes only information about processing components running on the node, for example.

  The table allows an overview to be maintained and allows quick access to the most relevant information regarding connections between multiple components. The identifier helps that the determination of crossing a node boundary can be performed faster and with fewer computing resources.

  The system, in particular the control unit that hosts the corresponding sending component and / or the execution environment, can be configured to detect if the communication channel crosses the boundary of the node by a search operation in the table. The search operation will detect whether the component has been moved, ie, the reassignment of the processing component, or whether an entry has been set that can indicate the creation of a connection with the component on the remote host. Can be configured. The entry may be a first flag set in the table. Alternatively, or in addition, a search operation can be configured to retrieve the identifier to find the location of the receiving component's node. The search operation allows the determination of whether the communication channel crosses the node boundary to be performed almost immediately without using too much computing resources.

  Further, for example, when a transmission and / or reception component is subsequently moved from one node to another when a communication channel is established between the two components, the system, particularly the execution environment, may It may be further configured to add the entry to the information within. An entry can specify component movement, which allows a faster determination of whether a communication channel, especially an existing communication channel, crosses a node boundary.

  Further, the system may be able to determine that encoding / decoding is necessary if the data to be transferred exceeds a predetermined size. In particular, the size or volume (network interface capacity) of the data stream at the transmitting or receiving side and / or the capacity limitation of the physical network channel are of interest. Typically, video related data or high quality audio data triggers the encoding / decoding requirements.

  The system may be able to determine that encoding / decoding is necessary if the data to be transferred has a predetermined data format, ie based on the format of the data type. The predetermined data format can include, for example, a media format. This is encoded when the system transfers data from one node to a remote node, typically data having media data, eg, image data format, audio data format, and / or video data format. It means that it can be configured to decrypt.

  Further, when the data is video data transmitted in units of frames, encoding / decoding may be necessary. When a component sends a complete frame of pixel values, i.e. an uncompressed video frame (raw data), to its output port, a large amount of data is exchanged between the components via a communication channel. This can make it difficult to divide application components across different processing nodes. However, automatic compression of the data described herein also allows video frames to be transmitted, thus enabling automatic distribution of video data application components.

  The system may be able to determine that encoding / decoding is necessary when one or both of the sending and receiving nodes operate at a given workload. This is because the encoding / decoding additional workload will lead to overloading at least one of the nodes when the node is already operating at a high workload. Cases where the encoding is not initialized can be included.

  According to one aspect, the transmission processing node is at least one execution environment, at least one processing component, and at least connectable via the communication channel to the reception processing node to transmit data via the communication channel. One outgoing port can be provided. The execution environment can be configured to establish a communication channel. The reception processing node can comprise a control unit. Further, the transmit processing node can be configured to instantiate an encoder and inform the receive processing node about the type of encoder instantiated when it is determined that encoding is required. Further, the receiving processing node can comprise at least one processing component and at least one incoming port connectable via the communication channel to the transmitting node for receiving data via the communication channel. The transmission processing node may comprise a control unit. The receiving processing node may be further configured to instantiate the decoder based on information regarding the type of encoder instantiated at the transmitting processing node when it is determined that encoding is required.

  Furthermore, a (processing) node may be provided that may comprise an execution environment and at least one processing component. The execution environment may include means for connecting the outgoing port of the component transmitting data with the incoming port of the component receiving data that may be located at different nodes via a communication channel. A node (eg, the control unit of the node) can be configured to determine if the communication channel crosses a node boundary. In addition, a node may encode data transferred over a communication channel to reduce traffic on an external communication channel, eg, a communication network connecting the nodes, when the communication channel crosses a node boundary. / Can be configured to determine if decoding is necessary.

  The node can be part of a network with further processing nodes. Nodes can be connected to each other in the network. The advantage of the node described above is that the node can automatically determine whether data should be transferred over the communication channel in an encoded manner. For example, assuming that a node is a unit having at least one processor, at least one storage means, and a network interface, the node communicates when a communication channel connects outside the unit via the interface. It can be determined that the channel crosses the boundary of the node. In other words, a node can determine that a node boundary is traversed when a communication channel connects the node's network interface to a different network interface of a different node. The node can further determine that encoding is necessary when the type of data transmitted via the communication channel is a specific type, for example, a media data type such as voice, image or video data. .

  Note that the functionality and aspects of the above system can be incorporated into each node. For example, the determination of whether a communication channel crosses a node boundary and whether encoding / decoding is required can be performed by either the transmitting or receiving node, particularly its execution environment. The table can be handled by the function of the node, and the search operation can be executed by the function of the transmission node or the like. Alternatively, a control instance of the entire network can be provided to implement these functions.

  In addition, the system can include an application distribution function that can automatically distribute the processing components of the application across the nodes so that each (processing) component can be processed by the node individually / separately. Typical applications relate to the processing of multimedia applications such as video chat, voice chat, and video streaming.

  Thus, the system communicates to transmit data such as video data or any other media data between service components whenever a network link crossing (ie processing node external communication between two components) is detected. It would be advantageous to be able to dynamically instantiate and include compression functionality in the channel data stream.

  For example, in a platform such as a multimedia content cloud, an application developer does not know when a message traverses a network link on a communication channel established between two components because the component placement decision depends on the platform This is because it is executed dynamically at runtime. In the proposed system, whenever a processing node boundary (ie, an external network interface or network link) is traversed by a communication channel and / or the expected data size or data rate can introduce overload on the network link, The execution environment instantiates the compression function (ie instantiates an encoder on the resource hosting the transmitting component and a decoder on the resource hosting the receiving component).

  In such a case, according to an embodiment, the execution environment determines the basic codec (coder and decoder) format and automatically instantiates the transmitting encoder and the receiving decoder as part of the message passing interface. . For example, in a platform scenario, the components can communicate via a message passing interface that connects the output port of the sending component to the input port of one or more receiving components. Codec instantiation, for example, when a new communication channel between a local component and a remote component is established, or the platform resource management algorithm decides to move one of the components to another processing resource / node This can happen if an existing local connection becomes a remote connection.

  The codec format passes along the message type of the incoming port of the receiving component or the outgoing port of the sending component, the specific (specified by the application developer) or connection of the logical link used to exchange data between the two components This can be determined by the type of message to be performed.

  When sending a message across a connection, ie communication channel, the platform function passes the data to the encoder, compresses it, and sends it over the external network using standard network protocols (eg RTP, UDP, TCP, IP) And then decrypt it remotely and finally pass it to the receiving component in its original uncompressed form. Data transmission is performed to facilitate distributed processing of applications. Processing of application components is performed by processing nodes connected to each other in the network.

  A further aspect relates to a method for data communication in a system. The system can have at least two nodes. A node can have an execution environment and processing components. The execution environment can include means for connecting the outgoing / outbound port of a processing component capable of transmitting data with the incoming port of the processing component capable of receiving data via a communication channel. The method may comprise determining whether the communication channel crosses a node boundary. This can be done, for example, when establishing a communication channel between the sending node and the receiving node. Further, the method can comprise determining whether encoding / decoding of data transferred via the communication channel is necessary, particularly if it is determined that a node boundary is crossed. The method may comprise initializing an encoder in the node of the transmitting component and a decoder in the node of the receiving component if it is determined that encoding / decoding is necessary. The method can further include providing the data to be transferred from the transmitting component to the encoder and encoding the data within the encoder. The encoded data can be transmitted to the receiving node via the communication channel.

  This method automatically processes the data transmitted through the communication channel so that the data traffic generated in a system in which the application is distributedly processed, that is, the application is processed by different processing nodes can be flexibly reduced. There is an advantage that encoding / decoding can be performed.

  Further, after the encoder is instantiated in the transmitting component node, a control message (information) that may include information on the type of the instantiated encoder is executed from the execution environment of the transmitting component node, etc. It can be sent to the environment. The decoder can be instantiated based on information included in the control message. Preferably, when the decoder is instantiated, the transferred data can be transmitted from the transmitting node to the receiving node via the communication channel. Alternatively, the control information can be transmitted along with the transferred data, for example, the transferred data can be transmitted in a message that also includes control information regarding the encoding / compression function applied at the transmitting side. By sending the control information and the data to be transferred in one message, the network traffic can be further reduced.

  Thus, there are several alternative options for instantiating codecs and for transmitting data via communication channels that cross node boundaries. After the encoder is instantiated at the sender, the control message can be sent to the receiving node. The control message submits information about the encoder type etc. to the receiver so that the corresponding decoder can be instantiated. A further option may be that the encoder can be instantiated and the transferred data can be encoded at the sender. According to this option, subsequently, a message including encoded data, for example, a message including control information regarding encoding / compression in the header portion of the message can be transferred to the reception side. Therefore, both control information and transferred data arrive at the receiving side at the same time. The decoder can be instantiated according to the control message in the message before decoding the transferred encoded data. Yet another option may be to use a predetermined codec that can be fixedly instantiated for the communication channel, such as when the communication channel is established. A given codec can be instantiated prior to sending the message / data via the communication channel. For example, the option of instantiating a given codec may be preferred if fixedly configured encoding / decoding is to be used for all data transferred over the communication channel .

  A second flag can be added to the data to be transferred, which is done, for example, by setting the second flag in the header of the data message when the data is compressed data. This can be done to indicate to the receiving component node that decoding is necessary so that the receiving side can quickly detect that decoding is necessary. Furthermore, the second flag can also be used to indicate to the receiving side the encoding type format used by the transmitting side encoder.

  The method may comprise the transmitting encoder and the receiving decoder being instantiated upon establishment of the communication channel and before transmitting data via the communication channel. Furthermore, the type of encoding / decoding is fixed for the communication channel based on the expected data type, eg audio or video data, and / or the size of the data transferred over the communication channel. Can be preset. Therefore, the developer can set a suitable codec format for a specific communication channel based on his / her knowledge about data to be transmitted via the communication channel. For example, a developer may specify a video codec in the expectation that only video data will be transferred. In this way, a fast setup time can be achieved.

  In summary, it is an advantage of the aspects presented herein that application developers do not have to work on detailed component control and still provide greater flexibility in application design. It is the platform, not the service developer, who is responsible for finding the optimal instantiation of components on the available resources. This reduces the development effort to build a distributed cloud-centric media service because developers no longer need to pay attention to the placement of components on resources that may even change during service execution.

  Note that the aspects outlined above can be implemented in a data processing apparatus comprising a processor. Computer program instructions stored in a computer memory can implement aspects when executed by a processor. Embodiments are further described below with reference to the accompanying drawings.

FIG. 6 is an example of a distributed processing component. 1 is an example of a pipeline-oriented media framework. FIG. FIG. 2 is a diagram of the main concepts of the disclosed system. FIG. 3 is an example of an automatically distributed processing component of an application.

  Embodiments conform to actual standards, cloud servers hosting distributable video and other multimedia applications, end devices such as mobile phones, network cameras and other entities capable of recording or displaying video and multimedia content Note that it can be technically implemented in an Internet gateway that serves as a connection point between a network device and a platform service incorporating the proposed encoding format.

  The basic integration of the proposed solution into a proprietary cloud environment should be possible without standardization, but a similar problem is moving mobile computing-intensive video processing tasks into a network-centric cloud environment. Since it applies in the environment when offloading, there is a potential adoption by ITU / ISO (MPEG).

  In a cloud-based processing environment, in particular, the following data transmission scenarios between separate software tasks running on separate processing nodes may be applied. 1) Media data transmission to end devices (eg, cameras, displays, smartphones) and from end devices to media processing nodes across networks such as the public Internet. 2) Media data transmission between processing nodes in a single data center. This setup is done because media processing is computationally complex (ie, advanced processing), so media processing tasks can easily scale out of a single processing node. 3) deployed in different data centers, for example to transmit resources within the network to optimize overall resource consumption, eg to place processing close to where the user is and to limit delays related to man-machine interaction Media data transmission between different processing nodes.

  This embodiment specifically addresses cases 2) and 3) and assumes that control of transmission between the end device and the cloud node is provided by having the ability to run dedicated software on the device, for example. It should be noted here that case 1) is also covered.

  For example, a typical application scenario that benefits from distributed processing on a service platform is, for example, video chat. In a video chat scenario, for example, a group of people participate in a virtual shared video environment / video chat room. Each individual streams their camera video data to the cloud. The video chat service combines all media streams of users gathering in such a shared environment and generates (ie, processes) a personal view of the virtual room for each participant. Thus, each individual in the virtual room receives a personalized rendered view of the other participants.

  1 and 2 are examples illustrating possible distribution of processing components 3 to a distributed computing node 1 including codecs (coder 5 and decoder 6). Specifically, FIG. 1 illustrates how service logic and resources can be divided into logically dependent functions, for example, for a video chat application. In a possible deployment, each of the illustrated service logic blocks can ultimately be a different resource.

  FIG. 2 shows a pipeline-oriented media framework with exemplary details of independent pipelines, each pipeline having a data receiver and encoder / decoder component at each possible transfer point. FIG. 2 specifically illustrates service logic instantiation including different application tasks such as components such as camera processing, virtual pan / tilt / zoom, background extraction, video blending, and personal view rendering.

  FIG. 3 shows the main design of the system (platform) according to the embodiment. In a preferred embodiment, the system includes at least two processing nodes 1 connected by a network such as an IP-based Internet, Ethernet. The processing component 3 of the application to be processed by the system is distributed to the two illustrated nodes 1 in FIG. 3, for example. In other words, the application is divided into a plurality of parts, for example processing components or sub-applications. The application distribution function distributes the component 3 to the node 1 of the system, here, for example, the first node 1a and the second node 1b. Data exchange between the nodes 1 is possible via the communication channel 4. Data transmission is preferably performed, for example, by passing a message from one node to another. The message can comprise a header and a payload or data to be transmitted.

  More specifically, a distributed processing platform is executed on each node 1 to instantiate, modify and delete the media processing component 3. The processing component (component) 3 is a processing part of an application divided into the different components 3. Each component 3 has a unique identifier across the platform.

  For example, a video chat application may pre-process camera signals such as background subtraction, face or behavior detection, eg “who is talking” functions, virtual cameras in real camera video signals where the user may be high resolution Can be divided into components such as an interactive module that generates a personal view that can be controlled by moving the pan, tilt, and zoom elements of the camera, and that can be of low resolution. In addition, the component has a video mixing function that merges the different video signals of people in the chat room on the processing resource in the cloud (preferably placed in “center of delay” for all participants), and Can be a personal viewpoint rendering (preferably located on a processing resource located near this user), controlled by the end user's interactive gestures.

  Specifically, FIG. 3 shows that each node 1 has an execution environment 2. The execution environment 2 comprises means for connecting at least two components 3 by instantiating a communication channel 4 between the outgoing or outbound port of the first component 3a and the incoming port of the second component 3b. Such a connection 4 can be realized in a distributed manner by establishing a service logical data flow and maintaining a table describing all communication channels 4 of the component 3 operating on the node.

  In an embodiment, a new connection between two components 3 is instantiated by adding an entry to the table. The entry indicates which subsequent component 3 is the receiving component 3 (and its port) of transfer data (eg, media stream) transmitted by the sending port of the sending component 3. The table can be controlled by the platform.

  Furthermore, if the receiving component is located at the remote node 1, this means that the communication channel 4 crosses the node boundary, but the platform will identify the remote host (node) identifier, which is the IP address. Get but add to table entry.

  FIG. 3 further illustrates two possible scenarios. The first scenario is shown in the upper part of FIG. 3, where the two components 3 of the application are both located on the same node 1. Accordingly, the execution environment 2 instantiates a local communication channel 4 between the two components 3 without instantiating the encoder 5 and / or the decoder 6. That is, the data (payload) transferred from the transmission component 3 a to the reception component 3 b is preferably transmitted without being encoded via the communication channel 4. Thereby, computing resources are maintained. For example, a message that is a header and data can be transferred by memory reference.

  In the lower part of FIG. 3, it is illustrated that the receiving component 3b is arranged in the node 1b and the node 1b is arranged remotely with respect to the other nodes 1a including the transmitting component 3a. Therefore, data transfer via the communication channel 4 requires a network protocol, for example. The reason for placing the two components 3 remotely is, for example, that the platform first placed the component 3 on a different node 1 or the receiving component 3b is moved from one node 1 to another node 1 during processing. Could be. An example of moving the component 3 to the remote node 1 after the communication channel 4 is established is shown in the upper part of FIG. The rearrangement of the receiving components 3a and 3b is indicated by the arrow “component movement”.

  In general, when data is to be transferred from the sending component 3a to the receiving component 3b, the sending component 3a calls a platform function that specifies a message type, a payload, and a sending port for the transferred data.

  Subsequently, the identifier of the receiver component 3b and its port are searched using a search operation on the table. The search operation is configured to determine if one or more recipients of the forwarded message are located at the remote host. If the message receiver is located on the same host as the transmitter, the message is stored in a queue inside the node and finally delivered to the target receiving component. If the message receiver is located at the remote host, the message is sent, stored in a queue on the remote node, and finally passed to the receiving component message handler. At this time, the port index of the transmitter is replaced with the port index of the receiver, and the two components 3 are completely separated. Thus, the sending component 3a does not know where the corresponding endpoint, ie the receiver of the message is located, because the only connection between both components is manipulated by the execution environment etc. and accessed by the component itself. This is because it is realized by information stored in a table that may not be possible.

  The platform (eg, the control unit of the system (not shown in FIG. 3)), from the search operation, the connection / communication channel 4 between the components 3 traverses the network link (the boundary between the two nodes 1). Recognize. Furthermore, specifically by analyzing the message type or payload size, it is possible to determine whether the data link can cause network link congestion. This aspect of automatic compression is further described below, particularly with respect to the lower portion of FIG.

  Initially, if the connection is established because the corresponding component 3 has already been instantiated on the remote processing resource, or the platform removes component 3 for other reasons, such as load balancing or other performance reasons. Note further that the connection can be external if it is decided to move to processing node 1 (see eg FIG. 3).

  According to one aspect, after the creation of the communication channel 4 between the two components 3 or after the movement of the component 3 to the remote node 1 (“component movement”), the first associated with the respective table entry. Is set. If the next message is sent over communication channel 4 and the first flag introduced is set, the platform parses the message header to determine the message type and the potential load on the network. And determine the appropriate encoder for the corresponding message type. The analysis of the message header is specifically performed in the following exemplary scenario: a communication channel 4 is already established between two components 3 located on the same node 1. At least one of the two components 3 is then moved to the remote node 1 and a first flag is set to indicate component movement. Before the next message is sent via the communication channel 4, the system sets the first flag set so that the communication channel 4 passes the border of node 1 and the payload data is encoded. Detect what may be needed.

  If the message is of a particular type for which a compression module is present and / or the data has a characteristic size, the platform (eg, the control unit of the system) may encode / decode to the communication channel 4 It can be determined that it is necessary. Furthermore, it is determined that encoding / decoding is necessary when one of the two nodes 1 connected by the communication channel 4 or at least one of the two nodes 1 operates with a specific workload. can do.

  A specific example may be that the system (eg the control unit of the system) always initializes the codec when the data is media data, eg video or audio format data.

  If encoding / decoding is required, and preferably when the codec is available with respect to the type of data being transferred, the corresponding encoder 5 is initialized and responds according to one option to instantiate the codec. A control message is transmitted to the receiving node 3 indicating that the decoder 6 that needs to be initialized with respect to the communication channel 4 on the remote resource (node) 1 that hosts the receiving component 3. The transferred (payload) data is transferred from the transmission component 3 to the encoder 5 and encoded. Subsequently, the encoded data is transmitted from the encoder 5 to the remote side. The message comprising the data to be transferred can further comprise a second flag indicating that the transferred data is compressed. In the node 1 including the receiving component 3, it can be confirmed whether the incoming / forwarding data is encoded. If the data is encoded, the decoder 6 is instantiated and the decoder 6 decodes the data. Thereafter, the decrypted data is transferred to the receiving component 3. Possible codecs are MPEG-1, MPEG-4, VPS, H.264. H.264, H.C. 265, MELP, SILK, and the like.

  As described above, the original data type can be maintained in order to appropriately transfer the transferred data to the receiving component 3. For example, the original message type message used for data exchange between processing components is encoded and encapsulated in a new message exchanged between the execution environments of the respective nodes. The receiving node's execution environment receives this message, passes the included encoded original payload data to the decoder, generates a message of the original message type, and forwards it to the receiving component. Thus, the data (message) encoding / decoding operation is transparent to the processing component. Optionally, control information can be inserted into a message that further includes the encoded forwarded data, including information about the required decoder 6 and indicating the need to instantiate the decoder 6. This avoids requiring additional communication between processing nodes 1 to ensure that encoder 6 is correctly initialized, such as sending a separate control message.

  According to a further aspect, the encoder 5 and the decoder 6 are initialized when a remote connection is established (even before the first message is sent). This can be done if the application developer knows in advance about the nature of the compression format and if this is indicated by a connection creation system call, or if the formal description of the component indicates the expected message type and This is the case when analyzed to determine the traffic rate. For example, since only video data is transferred via a specific communication channel, video codecs such as MPEG-4, VP8, H.264, etc. H.264 or H.264 The application developer may expect or know that 265 may be initialized for this particular communication channel. As a result, several alternative options for instantiating codecs are disclosed herein and are described in more detail above.

  FIG. 4 shows the result of a deployed application such as a video chat service, which is an optimized partitioning of the application and automatic encoding of the data, i.e., as described above, the data on different nodes By being able to automatically encode and exchange between ones, it uses as few resources as possible on the platform. For example, FIG. 4 shows that the pan / tilt / zoom (PTZ) component 3 of the video chat application is distributed to two different nodes 1 ′, 1 ″. One of the two nodes 1 'is a complex codec such as H.264. H.264 and video data having a low resolution, eg VGA. The other node 1 "processes the PTZ for video data having a high resolution, for example 720p, and a simple codec, for example MP4V. Both PTZ processing nodes 1 ′, 1 ″ can establish a communication channel 4 with a background extraction (BG) component located on yet another node 1 ″. Furthermore, both PTZ processing nodes 1 ′, 1 ″ in FIG. 4 can also send data to a node processing personal view (PV) rendering via the communication channel 4. In this example of FIG. 4, node 1 ″ ″ ″ is used to process high resolution PV rendering, and an additional node 1 ″ ″ ″ is used to process low resolution PV rendering. Since low resolution PV rendering does not require much computing resources, the platform will place the video mixing component 3 on the same node 1 ′ ″ ″ in this example of FIG. 4 as the low resolution PV rendering. Has been decided. In summary, FIG. 4 shows an example of an automatically distributed video chat application. Data transfer between the processing components 3 can be performed via the communication channel 4. For example, between PTZ processing nodes 1 ′, 1 ″ and BG processing nodes 1 ′ ″ or PTZ processing nodes 1 ′, 1 ″ and PV rendering processing nodes 1 ″ ″, 1 ″ ″. As shown in FIG. 4 for the communication channel 4 between ', the system can automatically instantiate the codec to reduce data traffic if the communication channel 4 crosses the border of node 1 .

  In summary, the proposed nodes, systems and methods allow for flexible deployment of media applications in a distributed (cloud-based) execution environment. Application developers do not need to remember resource constraints and can only manage service logic. This means that the cloud is virtually a single resource. By applying the proposed method and the system that implements this method, compared to a more granular method, such as a virtual machine or a method realized by deploying even dedicated service components introduced by application developers Thus, the platform can divide application service logic among computing resources even more efficiently, thus improving the overall resource utilization. This is thought to achieve a higher utilization of about 30-40% compared to existing solutions. Service and application development is simplified, resulting in wide adoption of the underlying cloud platform.

  While the above describes a particular sequence of operations performed by a particular embodiment, alternative embodiments perform the operations in a different order, combine certain operations, duplicate certain operations, etc. It should be understood that such an order is exemplary. References herein to a given embodiment indicate that the described embodiment can include a particular function, structure, or feature, but not all embodiments necessarily include a particular function, structure, or feature. Sometimes.

  Note that the description and drawings are merely illustrative of the principles of the proposed method and system. Accordingly, although not explicitly described or illustrated herein, it will be understood that various configurations may be devised by those skilled in the art. In addition, all examples listed herein are primarily educational to assist the reader in understanding the principles of the proposed method and system and the concepts devoted to the development of the art by the inventors. It is expressly intended to be for purposes only and should not be construed as limited to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments, as well as specific examples thereof, are intended to include equivalents thereof.

  It is further noted that the various method steps described above and the components of the described system can be implemented by a programmed computer. As used herein, some embodiments also encode a program of machine-readable or computer-readable and machine-executable or computer-executable instructions, said instructions being the steps of the above-described method. It is intended to include a program storage device, such as a digital data storage medium, that implements part or all. The program storage device can be, for example, a digital memory, a magnetic storage medium such as a magnetic disk and magnetic tape, a hard drive, or an optically readable digital data storage medium. Embodiments also include a computer programmed to perform the steps of the method described above.

  In addition, it should be noted that the functionality of the various elements described in this patent document can be provided using dedicated hardware as well as hardware capable of executing software in conjunction with appropriate software. These functions, when provided by a processor, can be provided by a single dedicated processor, a single shared processor, or multiple individual processors, some of which can be shared. Furthermore, the explicit use of the terms “processor” or “controller” should not be construed to refer exclusively to hardware capable of executing software, but implicitly, but not limited to, digital signals Processor (DSP) hardware, network processor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), and non-volatile A storage device may be included. Other hardware can be included as well, whether conventional and / or custom.

  Finally, it should be understood that any block diagram herein represents a conceptual diagram. Similarly, any flowchart, flowchart, state transition diagram, pseudocode, etc. can be substantially represented on a computer-readable medium, and thus such computer or processor is explicitly indicated by the computer or processor. Anyway, it will be understood that it represents various processes that can be performed.

Claims (15)

A control unit comprising an application distribution unit configured to automatically distribute processing components (3) of an application across processing nodes (1) of a distributed processing service platform ,
It is configured to determine if the communication channel (4) between the two processing components (3) crosses the boundary of the processing node (1), the communication channel (4) being one of the processing nodes (1) one of the case where it is determined to cross the border, which is further configured to encode and / or decode the data transferred via communication channel (4) to determine whether the required control unit. A system comprising at least two processing nodes (1),
At least one node (1) comprises an execution environment (2) and at least one processing component (3);
Said execution environment (2) comprises means for establishing a communication channel (4) between an outgoing port of a processing component (3) and an incoming port of a different processing component (3);
Said execution environment (2) is further configured to instantiate, modify and delete processing components (3) respectively distributed on node (1);
The system further comprises a control unit according to claim 1,
system. The system further comprises a table containing information on said established communication channel (4);
Wherein the information comprises a port of the transmission component (3) and receiving components of communication channels (3), said information, receiving component (3) is arranged to transmit the component (3) and different nodes (1) in Further includes an identifier,
The identifier identifies the network location of the receiving node (1);
The system according to claim 2. The system is configured to detect whether the communication channel (4) crosses the boundary of the node (1) by a search operation in the table;
A search operation detects whether an entry indicating reassignment of component (3) is set in the table and / or uses the identifier to find the location of node (1) of receiving component (3) Configured to read,
The system according to claim 2 or 3. 5. A system according to at least one of claims 2 to 4, further configured to determine that encoding / decoding is necessary if the transferred data exceeds a predetermined size. It is further configured to determine that encoding / decoding is necessary when the data to be transferred has a predetermined data type, and the predetermined data type is preferably a media data type, eg, image data, audio data, video 6. A system according to at least one of claims 2 to 5, which is data. 7. At least one of claims 2 to 6, further configured to determine that encoding / decoding is necessary when one or both of the sending and receiving nodes (1) operate at a given workload. System. The processing component (3) communicates via the message passing interface provided by the execution environment (2) of each node (1), and the message passing interface connects the output port of the sending component (3) to the receiving component (3). The system according to at least one of claims 2 to 7, wherein the system is connected to an input port . A transmission processing node (1),
At least one execution environment (2) configured to instantiate, modify and delete the distributed processing component (3) of the application on the node (1);
Comprising at least one distributed processing component (3) and operating on the node (1),
At least one outgoing port of the processing component (3) is, receiving processing nodes to transmit data via a communication channel (4) (1) and can be connected via a communication channel (4), the execution environment (2) is configured to establish a communication channel (4);
The transmission processing node (1) comprises a control unit according to claim 1,
A transmission processing node configured to instantiate the encoder (5) and inform the reception processing node (1) about the type of the instantiated encoder (5) if it is determined that encoding is necessary. A reception processing node (1),
At least one execution environment (2) configured to instantiate, modify and delete the distributed processing component (3) of the application on the node (1);
Comprising at least one distributed processing component (3) and operating on the node (1),
A communication channel (4) through a communication channel (4) Transmission processing node (1) for receiving data via a connectable,
The reception processing node (1) comprises a control unit according to claim 1,
Based on the information received from the transmission processing node (1) regarding the type of encoder (5) instantiated in the transmission processing node (1) when it is determined that encoding is necessary, the decoder ( A receive processing node configured to instantiate 6). A method for data communication in a system having at least two processing nodes (1), comprising:
At least one node (1) has an execution environment (2 ) for instantiating, modifying and deleting processing components (3) distributed on each node (1);
Said execution environment (2) comprises means for connecting an outgoing port of a processing component (3) for transmitting data with an incoming port of a processing component (3) for receiving data via a communication channel (4);
Automatically distributing the processing components of the application on node (1);
Establishing a communication channel (4) between the transmission processing component (3) and the reception processing component (3);
Determining whether the communication channel (4) crosses the boundary of the node (1);
If the boundary of the node (1) is determined to be traversed, the steps of the encoding / decoding of data transferred via communication channel (4) to determine whether it is necessary,
Initializing the encoder (5) in the node of the transmitting component (3) and the decoder (6) in the node of the receiving component (3) if it is determined that encoding / decoding is necessary; ,
Providing the data to be transferred from the transmission component (3) to the encoder (5) and encoding the data in the encoder (5);
Transmitting the encoded data to the receiving node via the communication channel (4). After the encoder (5) is instantiated by the node of the sending component (3), a message containing control information regarding the type of the instantiated encoder (5) is sent to the execution environment (2) of the node of the sending component (3). To the execution environment (2) of the node of the receiving component (3),
A decoder (6) is instantiated based on the control information included in the message.
The method of claim 11.   13. The method according to claim 12, wherein the data to be transferred is transmitted from the transmitting node to the receiving node (3) via the communication channel (4) when instantiating the decoder (6).   The method of claim 12, wherein the transferred data is transmitted in a message that also includes control information.   The transmitting encoder (5) and the receiving decoder (6) are instantiated simultaneously with the establishment of the communication channel (4) and before transmitting data via the communication channel (4). Item 15. The method according to at least one of Items 11 to 14.

Images